Investing & Dividends Mostly accurate, with one big caveat

By the Income Reality Check Editorial Team · Updated 2026-10-02 · 7 min read

This article is general information, not financial, tax, or investment advice. Income claims and platform fees change. Talk with a licensed professional before making financial decisions based on anything you read here.

An AI agent traded $10,000 in stocks for a week — and lost to the S&P 500

Verdict: Mostly accurate, with one big caveat. The video honestly shows the bot losing, but seven trading days tells you nothing about whether any of this works.

Nate Herk’s video “I Gave GPT 6 Astra $10,000 to Trade Stocks…And This Happened” has pulled in more than 95,000 views on his AI Automation channel. The premise is simple and the title is built to make you click: real money, a frontier AI model, seven days, one goal — beat the S&P 500. Here’s the part the thumbnail doesn’t shout. The bot lost. It finished at roughly $9,900, down about $100, and it lost to the index by around 0.8%.

That’s not a gotcha. The creator says it out loud. Which makes this a more interesting video to check than the usual hype reel, because the honesty in the narration hides a caveat that matters a lot more than the final dollar figure.

What the video actually claims

The setup is genuinely novel. Herk gives an AI agent running in Codex on “GPT-6 Astra” a $10,000 Alpaca brokerage account and a rulebook: it wakes up on a schedule — before the open to read news, an hour in to look for a trade, midday, afternoon, and twice near the close — and each wake-up leaves a handoff note so the next one picks up where the last left off. He’s allowed up to two manual tweaks a day. The benchmark is the S&P 500. Beat it and he wins; lose and he gives away a VIP event ticket.

He hits a wall on day two. The model refuses to place trades on its own (“I can help you strategize, but I’m not going to actually execute these”), so he bolts on a second bot — a “GrokBot” named Trader — that receives the strategy by email and fires the orders through Alpaca. So the real architecture is one model doing research and a second one pulling the trigger.

To his credit, Herk never promises riches. He says plainly: “I’m not trying to tell you I’m going to turn 10K into 15K in seven days. That’s just very unrealistic.” The goal he sets is modest — beat a benchmark that was itself barely moving. By the end, after a few flat days, a Fed rate hike, and a last-day rally that faded, the account closes down about a hundred bucks and behind the index. A deal’s a deal; he gives away the ticket.

What the method actually requires

So if the creator is honest about the result, where’s the caveat?

It’s in what a seven-day test can and can’t prove. Herk himself admits he “only got truly five days of trading” and that this isn’t “enough to really know if this is a viable strategy.” That’s the whole ballgame. A week of trading is noise, not signal. Over such a short window, the gap between the bot and the S&P is almost entirely random — the Fed’s surprise 25-basis-point hike mid-week swamped anything the agent decided. You could rerun this experiment ten times and get ten different finishes, none of which would tell you whether the AI has any edge.

The longer-horizon data is what the video never puts on screen. Beating the market consistently is the hardest job in finance, and the professionals mostly fail at it. According to S&P’s SPIVA scorecard, as reported by CNBC, roughly 65% of active U.S. large-cap fund managers underperformed the S&P 500 in 2024, and only about 14% beat it over the trailing ten years. These are full-time teams with Bloomberg terminals, research staff, and decades of experience. An AI agent prompted to “be more aggressive” is not starting from a stronger position than they are.

Then there’s the direction Herk says he wants to go next: wake the bot up every 30 minutes, strip out the risk filters, and eventually add options. That’s a march straight into day trading — and the day-trading odds are brutal. Kiplinger, summarizing the research, notes that only about 1% of day traders consistently make money and that the activity is “much more like gambling” than investing. Academic studies across multiple countries put the share of day traders who lose money between 70% and 97%. Faster trading and looser filters don’t fix a losing strategy; they just let it lose faster.

Is an “AI trading bot” even the real product here?

Worth separating two things. Building the automation — scheduled agents, handoff notes, an email-triggered execution bot wired to a brokerage API — is a real and reasonably impressive technical demo. That part is true. The leap is assuming the automation implies an investing edge. It doesn’t. A bot that places trades quickly is only as good as the strategy behind the trades, and nothing in seven flat days demonstrates a good one.

This distinction is exactly where U.S. regulators have been drawing a hard line. The SEC calls the practice of overselling AI capabilities “AI washing,” and it has brought cases. In March 2024 the agency charged two advisers, Delphia and Global Predictions, for false claims about using AI in their investment process; they paid $225,000 and $175,000 in penalties. Later that year the SEC charged Rimar Capital and its CEO for raising nearly $4 million from 45 investors by describing a platform as “AI-driven” when it wasn’t. To be clear, Herk isn’t selling a fund or soliciting money — he’s running a personal experiment on camera, which is a different thing entirely. But the regulatory backdrop tells you why “I let AI trade for me” clips deserve a skeptical read: the words “AI” and “trading” together have been doing a lot of unearned lifting lately, and U.S. readers especially should treat any product built on that pairing with care.

Who actually wins this game

Short term, almost nobody beats the index reliably — and that’s the point. The people who come out ahead in markets over years are overwhelmingly the ones who stop trying to trade and just buy a broad index fund and hold it. The SPIVA data is a decade-long argument for that. The handful who do beat the market tend not to persist; this year’s winner is rarely next year’s.

The people who win from videos like this one are a different group: creators. A dramatic title, real money on the line, and a frontier model make for a compelling series — and Herk openly floats extending it to 30, 60, 90 days or a full year “and obviously make some videos about it.” That’s a smart content strategy. It is not evidence of a trading strategy.

What you’d realistically earn

If you copied this exact setup with your own $10,000, the honest answer is: anything could happen over a week, and over a year you’d most likely end up near — probably a bit below — what a plain S&P 500 index fund would have returned, after you account for the drag of frequent trading and the time you poured in. The video’s own result (slightly worse than the index) is the representative outcome, not the unlucky one.

Compare the effort. Herk spent a week building agents, babysitting a second execution bot, making daily tweaks, and staring at the screen on the final day. An index investor spends a few minutes, pays an expense ratio often under 0.05%, and historically beats roughly 85% of active pros over a decade without touching anything. If the goal is returns, the boring option keeps winning. If the goal is a fun engineering project you can film, that’s a legitimate reason to do it — just don’t confuse the two.

Who this is (and isn’t) for

This makes sense as a hobby for someone who finds the automation genuinely interesting, has money they can afford to lose entirely, and treats the result as entertainment rather than a retirement plan. If you’ve got a few hours to burn and $500 you wouldn’t miss, wiring up an agent to a paper-trading account is a decent way to learn how these tools work.

It is not for anyone hoping to “put AI to work” and collect passive income, anyone trading money they need, or anyone who’ll be tempted — as the video’s own arc suggests — to keep cranking up the aggression and add options after a flat week. That path has a well-documented destination, and it isn’t a bigger account.

What to remember

The video is refreshingly honest about its own outcome, and that honesty is worth crediting. The caveat is that a seven-day test proves nothing, the “get more aggressive” plan points toward the worst-odds corner of investing, and a slick AI setup is not the same as an investing edge. Building the bot is the real skill on display here. Beating the market still isn’t.

For a look at the strategy that actually clears that bar for most people, see our pieces on a simple buy-and-hold ETF portfolio and on realistic ways to make money with AI without a trading gamble.

Views and figures were accurate at the time of review and may have changed since publication.